Application of the convolutional neural network for recognition of the metal alloys microstructure constituents based on their morphological characteristics

This work details the results of the application of the deep learning approach to recognizing the morphological forms specific for particular phase constituents of alloy microstructure visible on microscope images. The elaborated procedure of the Neural Network Learning on this stage of the examinations gives an acceptable effectiveness of recognition of several morphological categories as sphere, polyhedron, petal, needle, ch. script, twig, dendrite based on recorded sets of microscope images of microstructure of chosen cast Al and Fe alloys. The main advantage of the Deep Learning approach presented in this work is the opportunity of its application to any collection of microscope images according to chosen defined classes. Simultaneously, the problems of using fuzzy logic to discriminate very similar objects can be avoided (as twigs and ch. script). Thus, the deep learning approach presented in this paper is a new and perspective tool for analyzing and discovering the knowledge contained in microscopic images of the polyphase alloy microstructure.